India Blockchain Alliance

Scientists from OpenAl, Google DeepMind, Anthropic and Meta have abandoned their fierce corporate rivalry to issue a joint warning about Al safety. More than 40 researchers published a research paper today arguing that a brief window to monitor Al reasoning could close forever - and soon.

Imagine waking up to find out you're running a business in another state. Except you've never left your city. Vilas Prabhakar Lad, a 51-year-old shopkeeper from Mumbai, lived this nightmare. What followed was five years of silence, shrugging, and sheer state and police apathy.

𝐋𝐋𝐌𝐬 𝐚𝐫𝐞 𝐚 𝐃𝐞𝐚𝐝-𝐄𝐧𝐝 𝐁𝐲 𝐃𝐞𝐬𝐢𝐠𝐧 𝐭𝐨 𝐄𝐯𝐞𝐫 𝐀𝐜𝐡𝐢𝐞𝐯𝐞 𝐑𝐞𝐚𝐥 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞. No amount of scaling will change that. You can spend trillion$ on GPUs, electricity, water, land, and monopolized compute, LLMs will still remain what they are today with all the same limitations. This isn’t ambition or innovation, it's pure insanity, repeating the same architectural mistakes, only bigger, and still expecting a different outcome. OpenAI now wants 100 million GPUs, yes, one hundred million. A $3 trillion infrastructure spend, consuming more power than entire countries, and infrastructure buildout that dwarfs all of cloud computing. All this, to push more tokens through a slightly bigger, slightly faster, still cognitively hollow architecture. We don’t download Wikipedia at birth. We grow through feedback, context, and experience. And we do it on 20 watts, not 20 gigawatts. Intelligence is what it is capable of becoming shaped by continuous interaction, reflection and autonomy not just what it does through brute-force pretraining. How is this the future of AI? It’s not. LLMs perform intelligence, not possess it. No matter how much funding, data or compute you throw at them, that won’t change. That’s not a scaling issue, it’s a design flaw. 𝐀 𝟏𝟎𝟎 𝐦𝐢𝐥𝐥𝐢𝐨𝐧 𝐜𝐚𝐧𝐝𝐥𝐞𝐬 𝐬𝐭𝐢𝐥𝐥 𝐰𝐨𝐧’𝐭 𝐠𝐞𝐭 𝐲𝐨𝐮 𝐭𝐨 𝐚 𝐥𝐢𝐠𝐡𝐭 𝐛𝐮𝐥𝐛. Yet, LLMs can be useful. They autocomplete, summarize, and assist with routine tasks under constant supervision. A whole industry has formed around that utility. But fluency isn’t intelligence and mistaking performance for cognition is the original sin of the LLM era. Not to be left behind, world's top AI labs from Meta, xAI, DeepMind etc fuel the illusion of intelligence by burning trillions on LLMs, insanity at scale. Yet, the AI leaders today openly say that you can summon LLMs to cure cancer, design a new chip, or reinvent civilization itself. And as they fall short and they will, they don’t rethink the approach or the architecture. They just train a bigger one, add more GPUs, drain the grid. Sacrifice privacy, agency and justify it all in the name of progress against the same old flawed architecture. Why are investors rewarding scale when it’s now obvious we’ve clearly hit an architectural wall? Why are they blatantly burning the planet for a prompt? Where’s the accountability for the power consumed, water used, the emissions generated just to hallucinate at scale? Where are the real cognitive breakthroughs? Where is the shift to a new architecture? Instead, why are we getting more compute, more hype, more synthetic benchmarks, more conference demos and performance theater at planetary scale? They say this is progress, but it looks like a desperate attempt to mask stagnated technology with sheer size. What we need now is a new foundation. One rooted in cognition, not brute force. One that learns, reasons, adapts, and grows. One that mirrors the way our minds work. 𝐑𝐞𝐣𝐞𝐜𝐭 𝐭𝐡𝐞 𝐈𝐧𝐬𝐚𝐧𝐞 𝐋𝐋𝐌 𝐈𝐥𝐥𝐮𝐬𝐢𝐨𝐧. 𝐓𝐢𝐦𝐞 𝐭𝐨 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐨𝐧. I know I’m in the less than 1% who believe LLMs aren’t on a path to real intelligence. But even if you’re among the 99% who do, you should be alarmed by the scale, by the spend and the thoughtless momentum behind it all. Speed only matters when you're headed in the right direction. Right now, we are moving fast in the wrong direction. No amount of prompt engineering or context engineering can fix a fundamentally flawed architecture. 𝐓𝐡𝐞 𝐑𝐢𝐠𝐡𝐭 𝐖𝐚𝐲 𝐭𝐨 𝐀𝐆𝐈—𝐀𝐟𝐭𝐞𝐫 𝐭𝐡𝐞 𝐋𝐋𝐌𝐬 is the blueprint and it's time we start engineering cognition.